For the complete documentation index, see llms.txt. Markdown versions of all pages are available by appending .md to any URL (e.g. /get-started.md).

Python class

ModuleV3SingleReplicaBatchProcessor

`ModuleV3SingleReplicaBatchProcessor`

class max.pipelines.lib.interfaces.ModuleV3SingleReplicaBatchProcessor(config, runtime, inputs=())

source

Bases: BatchProcessor[ContextT, InputsT, ArchConfig]

Single-replica ragged KV batching for ModuleV3 compile-path models.

Parameters:

get_symbolic_inputs(*, kv_params, device_refs)

source

Returns symbolic graph inputs for single-replica ModuleV3 models.

Parameters:

Return type:

list[TensorType | BufferType]

prepare_initial_token_inputs(replica_batches, kv_cache_inputs=None, return_n_logits=1)

source

Prepares ragged token inputs for a single-replica ModuleV3 batch.

Appends the per-call ModuleV3 LoRA buffers when the runtime’s manager is a LoRAManagerV3, reusing the row offsets already built here.

Parameters:

Return type:

InputsT

process_outputs(outputs)

source

Maps raw execution buffers to ModelOutputs.